DETECTION OF PARKINSON’S DISEASE

USING MACHINE LEARNING APPROACHES

Mark Emmanuel (Student), Chirag Khandelwal (Student), Aryan Bhanot (Student), Mamta Arora (Associate Professor)
Department of CST, Manav Rachna University

Abstract Methods and Materials

In recent years, there has been a substantial evolution of voice The research methodology involves utilizing voice recordings as a
technology, not only in itself, but also using it as a biomarker for biomarker for early detection of Parkinson's Disease through
medical and research purposes has proven its own benefits. We machine learning approaches. The study focuses on a 195-record
aspired to use this biomarker as a medium of detection of dataset from the University of Oxford, employing machine
Parkinson’s Disease. Parkinson’s Disease is a neurological, chronic learning algorithms such as SVM, Logistic Regression, Random
disease which progressively decays the body of the patient. As it Forest, Adaboost, and XGBoost. Data pre-processing includes
is a disease which usually shows its symptoms late in human life, addressing null values, correlation analysis, and attribute
its detection becomes crucial. Although there is no cure to the selection. Evaluation metrics, including accuracy, precision, recall,
disease, once detected early patients can search for alternative and F1 score, assess algorithmic performance. The research
medicines and treatment, and can even delay the symptoms by explores ethical considerations, data privacy, and responsible
the early diagnosis. In this research paper we have discussed the machine learning in healthcare. The aim is to contribute insights
use of voice as a biomarker for suspected Parkinson’s patients to into voice-based Parkinson's Disease detection and create a
detect any presence, or early symptoms in the form of any prediction model accessible to a larger audience.
speech impediments in the subject.
Chart 1. Research Plan

Introduction Results Requirements Resources

Parkinson’s Disease is a neurological chronic, and progressive After the implementation of the machine learning models on the For our project, we leveraged Google Colab for a Python-based
disease. With its conception in a person, it creates issues with the dataset. We decided to find out the best model to go forward coding environment, taking advantage of its seamless integration
simple motor abilities of the person. It causes problems with with for our research. Hence the following table shows the with popular libraries such as Scikit-learn, TensorFlow, and
movement, mental health, sleep, pain, amongst other health comparative analysis of all the algorithms that we tested on our Pandas. This choice allowed efficient collaboration and access to
issues. dataset. high computational power without the need for extensive local
As it is progressive once diagnosed it only worsens with time. At With careful consideration and discussion, we decided to choose hardware. Our dataset, sourced from the University of Oxford
present, there is no scientific actual cure to this disease, making it XGBoost as the machine learning algorithm that would be best fit repository, consisted of biomedical voice measurements in a
even worse on the immediate people(family) around the patient. for our problem statement of the detection of Parkinson’s structured format. The project team included college students
Disease. with a focus on machine learning. We prioritized ethical
Early detection and diagnosis of this disease can help in seeking
XGBoost stands for eXtreme Gradient Boosting. Gradient considerations, ensuring data privacy compliance, implementing
preventive measures early on in life. As this disease usually shows
Boosting is again a boosting technique that makes a final model bias mitigation measures, and adhering to responsible AI
its symptoms in older people, its detection at a younger age can
from the sum of several weak learning algorithms. XGBoost is a practices.
help in delaying the symptoms and easing it out for the mental
machine learning library that utilizes the gradient boosted
health of the patient and their family beforehand.
decision trees. The only difference between the two is that
Our research was done keeping all the above facts in mind, to
help and contribute in the field of medicine and sciences, in the
XGBoost applies a regularization approach too.ie. it gives a
regularized version of the prior gradient boosting technique. This
Conclusions
early prediction and detection of Parkinson’s disease. We wish to also makes it much more efficient on datasets that contain both In wrapping up our study, the outcomes affirm the potential of
create a prediction model which will take voice samples and the numerical and categorical variables. using voice recordings for early detection of Parkinson's Disease.
necessary attributes of the voice of the subject and assess Notably, the Random Forest and XGBoost algorithms
whether the person has a probability of being diagnosed with demonstrated high accuracy rates of 94.80% and 94.87%,
Parkinson’s Disease or not, using Machine Learning algorithms. Table 1. Comparative Analysis of Algorithms underlining their effectiveness in identifying potential cases. The
Further on, after the completion of this research we desire to F1 precision scores, reflecting the accuracy of positive predictions,
ML ACCURACY PRECISION
Heading RECALL CONFUSIO
Heading SCORE Heading
improve the accuracy of our model and make it accessible to the MODELS N MATRIX were commendable for both Random Forest (94.11%) and
larger audience, through creating a platform on which the SUPPORT [[ 5 3] XGBoost (94.11%).
Item 800 790 4001
individual can check their probability of Parkinson’s Disease. VECTOR 88.46 89.51 95.68 92.5 [ 2 29]]
MACHINE Across all models, XGBoost and Logistic Regression achieve
Item 356 856 290
LOGISTIC [[ 3 4] perfect recall scores of 100%. The F1 score, which balances
REGRESSION
Item 89.74228 88.88 100
134 94.1 238
[ 0 32]] precision and recall, further validated the overall efficacy of our
models, particularly with Random Forest and XGBoost exhibiting
Item 954 875
1 976 F1 scores of 96.96%.
RANDOM [[ 5 2]
FORREST 94.80324 94.11 100 96.9 [ 0 32]]
Item 325 301 These results highlight the potential of voice-based early
6 Parkinson's Disease detection, with Random Forest and XGBoost
Item
ADABOOST 199 137 186
[[ 4 3] showing promise for practical applications.
Figure 1. Physical problem in Figure 2. PD symptoms occur
walking in hands
ALGORITHM 87.17 90.90 93.75 92.3 [ 2 30]]

0
XGBOOST [[ 5 2]
Contact ALGORITH 94.87 94.11 100 96.9 [ 0 32]]
M
Aryan Bhanot Mark Emmanuel Chirag Khandelwal 6
Manav Rachna University Manav Rachna University Manav Rachna University
Email:aryanbhanot21@gmail.com Email:markemmanuelpc2004@gmail.com Email:khandelwalchirag2004@gmail.com
Phone: 9354512997 Phone: 9810578162 Phone: 9354638181